Contacts

Contacts is the address book at the centre of Unyo. It lives in your Neural Core, which means it isn't Maya's contact list or Alex's contact list — all 10 agents read the same one, on every run. Tell Ashley who your photographer is, and Maya knows their email the next time you ask her to write to them.

You'll find it on the People card in the Neural Core, reachable from the sidebar.

What a contact holds

Each contact is a structured record — not a line of prose an AI has to re-read and guess at:

Contacts also get an avatar colour on import, live in nested folders you can colour and rename, and can be selected in bulk.

Four ways contacts get in

1. Just say it

Type the fact to any agent, in any language, with no command and no prefix. The extractor runs after your message, decides on its own whether to create, update or merge against what's already there, and writes the contact.

You can also ask an agent to do it explicitly, which is the reliable route when you want certainty:

2. Import a file

Drop a file on the Contacts card and Unyo parses it into real contacts:

1. Open the Neural Core from the sidebar and click the Contacts card.
2. Drag your file in — or click to browse. Accepted: .csv, .xlsx, .xls, .ods, .tsv, .txt, .md, .pdf, .docx. Up to 50 MB.
3. Unyo detects the separator itself — semicolon, comma or tab — by reading your first five lines. No "which delimiter?" dialog.
4. Column headers are fuzzy-matched, so Nom, E-Mail, Teléfono, Firma, Şirket, 名前, 姓名, имя and الاسم الكامل all land in the right field. 10+ languages, including Arabic, Japanese, Chinese and Russian.
5. A folder named after your import is created automatically so the batch stays together.
6. Review what landed. Fix anything the file got wrong by clicking the contact.

If the file isn't a spreadsheet — a PDF member list, a .docx of attendees — the model reads it and extracts name, email, phone, role, company and a description instead. Same card, same drop.

3. Maya captures them as she emails

Send an email through Maya and the recipient is captured as a contact automatically. You don't ask for it and you don't confirm it — the address you just used is an address you'll use again.

This path is deterministic: no model decides whether the recipient "seemed important". It matches against your existing contacts and either updates the one it found or creates one.

4. Shopify customers sync in

Connect Shopify and your customers arrive in Contacts automatically, alongside your products and orders.

Managing contacts by talking

Every agent can read your contacts. Any agent that isn't restricted can also write them. Ask in plain language:

Updates only add and change what you named — an update never blanks a field you didn't mention. And exports go up to 200 entries as CSV or JSON, ready to hand on to Drive or Notion.

Privacy: the lock is real

Open any contact and mark it Private with the lock. That isn't a UI badge. A private entry is handed to every agent with an explicit instruction attached: do not mention this in any generated content, email or post. The agent can still use it to work; it won't leak it into a draft.

Use it for the contacts you'd rather not see quoted in a post Ashley drafts.

Deleting a contact

Delete a contact from the People card and it stays deleted — you can also ask an agent to delete one. The usual failure mode of automatic memory is that it re-learns what you just erased, because the conversation that taught it is still there. Unyo writes a tombstone on delete and feeds it back to the extractor as do not re-extract, so the contact you removed does not quietly reappear tomorrow.

Search

One box searches everything, and it looks inside your contacts, not just at their names — emails, phone numbers, company, role and notes are all matched. Searching northwind finds the person even if you only ever typed their company into the notes.

What Contacts costs

Nothing. Every Neural Core tool — read, search, create, update, delete, export — is priced at zero credits. Reading and writing your own memory is not a metered upsell. You're only billed for the model run itself, at real cost.

Tips

• Feed Contacts before you ask for outreach. Alex and Maya stop asking you for addresses the moment the address is already there.
• Import first, correct after. It's faster to drop the export and fix three rows than to clean the file.
• Say a preference once — "Mr Berger prefers WhatsApp and never answers before 10am" — and every agent honours it from then on. Preferences are shared across agents, not buried in a field.

Troubleshooting

A contact card looks empty after an import. Open it and check the Name field. Contacts need a real name of at least two characters to be created — rows with only an email may be skipped.

Something I said in chat wasn't saved. Automatic extraction has a confidence floor and captures a few items per message — it's deliberately conservative rather than eager, so it doesn't fill your memory with guesses. If a fact matters, ask for it directly: "Add a contact: …". That path is explicit and doesn't rely on the extractor's judgement.

An agent says it can't delete a contact. It's probably Shopify-synced — deletion of synced entries is refused in code. Remove the customer in Shopify and the change follows on the next sync.

My import put everything in one field. Check that your file has a header row — header matching is what routes columns into fields.